Special Guest Q&A: Get your questions in

More views on the interview than on the announcement, despite being almost one and a half hours vs a few minutes. Wouldn't have thought that. Congrats!

Phil

Question for Jim: In regards to Moore's Law. What instead of making the transistors smaller, why don't you make the surface bigger? Would like to hear Jim's explanation and perhaps experience with this.

M Timmers

Godspeed. I don't envy you on that one.

Thanks for the questions y'all, there's some good ones in here :) Apologies if I don't call your name out in the interview, I'm mixing and blurring what people are requesting. Now I have 500+ comments to sift through on the channel announcing the interview!

TechTechPotato

Would be great to learn more about his time at AMD. In addition to Zen, didn't he also prototype something ARM-based as well?

Thank you for the interview, Ian and Jim- What is the most common misconception that enthusiasts (industry 'outsiders') have about high level operations at semiconductor companies, or what do enthusiasts usually get wrong in their takes about the happenings at these companies?

Matt (CodgerFace)

First off, huge fan. Second, I'm not exactly clear how tenstorrent will change deep learning. From your previous interviews I gathered it had something to do with graphs. From a developer standpoint. How different will it be to code for your new product. Will we still be able to use pytorch and tensorflow? Sorry if this question doesn't make much sense. I just started getting into ml/dl myself.

Considering that there is an arbitrary code execution vulnerability in the universal turing machine https://www.theregister.com/2021/05/11/turing_machine_0day_no_patch_available/ as well as vulnerabilities in most processors that perform speculative execution, I want to know what Jim thinks about the future of cyber security in computing systems (especially as the hardware itself has become unfathomably complex). Also, how soon/does he ever think there will be dedicated hardware inside actual processor/accelerator dies to assist in mitigating vulnerabilities to either exfiltrate sensitive data about another process (like in spectre/meltdown) or inject data into another process (I am thinking of something more in the spirit of rowhammer here).

What are the major differences in management philosophy between AMD, Intel and Apple? Which parts of it come from the company culture vs immediate management chain? What is a neat feature in a CPU you worked on that people typically aren't aware of?

panic582d

What was the line of through behind the MCM design on ryzen versus a monolithic design? Was it purely cost/yield reasons or was there something else?

Questions for Jim: 1. Do you see a ternary/trinary or other multistate computing ever happening? 2. What will it take to make truly secure speculative execution? Are variable length instruction sets/microcode a barrier with regards to timing attacks? 3. What aspects of computing or specific algorithms would benefit from hardware acceleration that doesn't exist today? 4. In a world of ever increasing data, bandwidth has not kept up. What are your thoughts on moving computation to RAM? In other words, self-contained tile computing. 5. Are you afraid of AI becoming self-aware?

Mark Rose

What aspects of chip design and/or manufacture are most overlooked or least understood? Are there any aspects of the design and manufacture process that would be useful or helpful for the general public to have a better understanding of? E.g. to understand the current shortages, the length of development timelines, why 10ghz CPUs aren't around, or anything else that comes to mind.

Fawlters

How many Mars Bars are you comfortable with eating in one sitting?

with RISC cores in an x86 wrapper/translator, OSes with runtime executable libraries and interfaces (java, flash, dot-net, whatever), how many layers of abstraction can a system have before it becomes unruly to maintain and design. have neural-net, semi-AI, generational or genetic algorithm, etc, been used to design processors, branch prediction, optimize L1 cache/register sizes, etc? thoughts? or is it still mostly plotting-sheets, slide rules, and resistor paper on a far larger scale?

ZarconDeeGrissom

When, if at all, should we expect x86 to get extensions that enable more data centric computing techniques to be implemented in microarchitectures, particularly in and near memory processing? What similar forward looking technologies should we expect to see soon?

qhfreddy

Hi, Ian, Keep up the good work :) Here are few question that I could come up with: *What type of neuron model the company is basing their architecture on or is it a generic (Von Neumann) architecture processor optimized from machine learning algorithms? (at least high level) *What learning algorithm they are using or is it user dependent? (at least high level) *Could he talk a little bit about the process flow of the development of the processor? *Do they suffer with the fabrication bottle neck as well? *What process node are they using for their processor? *Have they tested their system with SNN (spiking neural networks)? *Have they compared the performance with neuromorphic processors such as Intel's "loihi" or IBM's "TrueNorth"? (Also a topic for future video) *How big is the company? *What is going to be his role in the company? Most of them sound very proprietary. But maybe he could answer some of them.

Hey Jim, what do you think the future of x86 architecture looks like in 3 years I.e. 2024

Andrew

Are variable length instructions in the x86-64 ISA a fundamental blocker for low power designs? Large, complected decoder consumes too much power or area budget?

What's your take on Intel vs AMD's rivalry over the next two years (consumer and enterprise)?

Kevin Wise